MAGNETIC RESONANCE IMAGING AIDED DIAGNOSIS OF OCCULT CERVICAL LYMPH NODE METASTASIS IN ORAL CANCER BASED ON ARTIFICIAL INTELLIGENCE TECHNOLOGY

This paper aimed to explore the application effect of magnetic resonance imaging (MRI) analysis technology based on a convolutional neural network (CNN) in the diagnosis of occult cervical lymph node metastasis in oral cancer. This study was a retrospective diagnostic investigation, enrolling a total of 240 patients with pathologically confirmed oral cancer. The patients were chronologically divided into a development cohort ([Formula: see text], used for model training and internal cross-validation) and an independent validation cohort ([Formula: see text], used for external performance evaluation). An improved Faster R-CNN detection framework that integrated features from T2-weighted imaging (T2WI) and contrast-enhanced T1-weighted imaging (CE-T1WI) was constructed. The model was optimized using transfer learning and a four-step alternating training strategy. The diagnostic performance of this artificial intelligence (AI) model was compared with that of the conventional raw data approach (RD, based on morphological measurements and radiomics scores). A multi-dimensional evaluation was conducted using the area under the curve (AUC), F1-score, Brier score, and the Hosmer–Lemeshow test. The short diameter, long diameter, and aspect ratio of lymph nodes in the metastasis group were visibly greater than those in the nonmetastasis group ([Formula: see text]). The radiomics scores of T2-weighted imaging (T2WI) and enhanced T1-weighted imaging (T1WI) images in the metastasis group were also visibly higher than those in the nonmetastasis group ([Formula: see text]). The MRI cervical lymph node metastasis diagnosis based on CNN was visibly superior to the raw data (RD) method in terms of sensitivity, specificity, accuracy, positive predictive value, and negative predictive value ([Formula: see text]). This paper indicates that the MRI cervical lymph node metastasis diagnosis method based on DNN is visibly superior to traditional methods in multiple key performance indicators, providing strong support for the early detection and precise treatment of occult cervical lymph node metastasis in oral cancer.